Search

This paper presents a new method for generating artworks that extends the classical anamorphic archetype to use freeform reflective and refractive media and 3D surfaces instead of images. The methodology uses a mix of raytracing and surface deformation techniques to determine the proper deformation the object should undergo to be corrected by the optical tool once viewed by the observer in a specific location. Our approach also includes an optimization process that modifies the point of view and the location of the target image to avoid unwanted optical effects or occlusions. We successfully tested our technique on several practical examples and employed it to produce actual artworks. • A new method to construct anamorphic sculptures using general optical media. • Characterized the main problems arising in general anamorphosis. • A simple system to allow the artist to create anamorphic art installations. • We created real-world art installations in international exhibitions. [ABSTRACT FROM AUTHOR]

This paper introduces a two-stage approach that automatically generates 3D CAD models from 2D orthographic drawings. First, a pattern-matching algorithm is proposed to reconstruct a network of 3D edges by matching 2D edge features extracted from the multiple views of the 2D drawing. Second, starting from the resulting 3D wireframe and generated graph, a loop detection algorithm allows identifying possible loops of faces. Then, a clustering algorithm recognizes the faces from the set of detected loops. The reconstruction ends while trimming and stitching the faces to get a watertight ready-to-use 3D CAD model. This approach has been validated on a public dataset composed of several thousands of 3D shapes, and it achieved 99.59% of well-reconstructed models in F-score. • A rule-based approach to automatically reconstruct 3D CAD models from 2D drawings. • A pattern-matching algorithm to generate 3D wireframes from 2D drawings. • A loop detection algorithm using co-planarity criterion to reduce the search space. • A clustering algorithm to identify the faces from a set of 3D loops in a wireframe. [ABSTRACT FROM AUTHOR]

Particle-based methods are practical for computing level-sets that represent liquid interfaces. However, these methods are computationally expensive to reconstruct the liquid surface when the number of particles increases considerably due to the massive amount of particle approximations. This paper introduces two simple and efficient surface reconstruction methods for particle-based fluids based on discrete indicator functions (DIFs). The first approach provides fast level-set approximation using a DIF defined by counting particles inside grid cells. The second approach generates a high-quality liquid surface using a DIF obtained by the particle distribution inside grid cells. Our DIF-based methods are fast, easy to code, and can be adapted straightforwardly in particle-based fluid solvers, even implemented in GPU. Moreover, we show the effectiveness of our approaches through experiments against prior surface reconstruction methods. [Display omitted] • A novel surface reconstruction method for meshfree liquid simulations based on Discrete Indicator Functions (DIFs). • We propose two novel DIFs that use only the particles inside the grid cells. • A scheme to recover small-scale details in surface reconstruction using union Boolean operation on DIFs. • Our framework speeds up considerably the surface reconstruction compared with prior methods. • Our methods are simple and easy to code, even in GPU. [ABSTRACT FROM AUTHOR]

Cryptocurrencies are a novel phenomenon in the finance world that is gaining more attention from the general public, banks, investors, and academic research lately. A characteristic of cryptocurrencies is to be the target of investments that, due to the volatility of most of the cryptocurrencies, tends to be at high risk and behave very differently from traditional currencies. A way of reducing this risk is to look at the history of existing cryptocurrencies and compare them to spot promising trends for increased gain. This paper introduces CryptoComparator, a Visual Analytics tool designed to analyze the correlations and trends of cryptocurrencies. The system exploits an initial proposal for a double elliptic graph layout, reconfigurable with three different ordering functions, to support a fast visual search of cryptocurrencies by correlation strength. Two use cases developed with a domain expert in cryptocurrency financial activities demonstrate qualitatively the support it provides for analyzing cryptocurrencies. [Display omitted] • A visual analytics solution for interactive analysis of cryptocurrencies. • Novel double ellipse graph layout. • Different graph orderings for correlation-based grouping of cryptocurrencies. • Correlation stability score to help the user in decision-making. • Two use cases on helping reducing volatility and building a crypto-wallet. [ABSTRACT FROM AUTHOR]

To capture the real-world luminance values, High Dynamic Range (HDR) image processing has been developed. HDR images have a richer content than the widely-used Standard Dynamic Range (SDR) images, and are used in a number of situations, e.g. in film industry. As HDR displays are more and more commercially available, we need to be able to process HDR images as well as SDR ones (for example, devising denoising algorithms, inpainting or anti-aliasing). The most powerful methods to process images are now deep neural networks. However, the training of such networks calls for a lot of images, and HDR images datasets are relatively small. One way to generate HDR images is inverse tone mapping operators (iTMOs). They are algorithms that expand the dynamic range of SDR images. In this paper, we propose HDR-LFNet, a novel iTMO, and its HDR training dataset. Our method merges several existing handcrafted iTMOs, combined in a supervised neural network to produce an HDR output. Our lightweight network requires less training images than state-of-the-art methods, and has faster training phase. Besides, the quality of the generated images is similar to the state-of-the-art. We present the architecture as well as the subjective and experimental evaluations of our method. [Display omitted] • We propose a new Inverse Tone Mapping Operator to create High Dynamic Range images. • Convolutional Neural Network are used to fuse together existing methods. • We performed a subjective study to evaluate our method. • The network's weights and training set are available online. [ABSTRACT FROM AUTHOR]

Omnidirectional images and videos are widespread due to the popularization of devices for capturing and visualization. Unlike pinhole-based imagery, omnidirectional media lie on the surface of a sphere, have a 360° × 180°field of view, and store the light intensities from an entire environment. Notably, applications involving immersive augmented, mixed, and virtual reality experiences benefit from the 360° content. Although defined on the spherical domain, omnidirectional images are often mapped to a (multi-)planar representation, which results in distorted images and degrades the performance of most traditional visual computing algorithms designed to work on the plane. This paper reviews the spherical camera model, the most common capturing devices, and popular (multi-)planar representations of omnidirectional media. It also approaches the main challenges of omnidirectional visual computing, focusing on the deep learning paradigm, and tackles four visual computing applications that strongly explore the potential of omnidirectional imagery. [Display omitted] • We review the foundations of the spherical camera model. • We review standard 360°capturing devices and format representations. • We discuss the main challenges of 360°visual computing and possible solutions. • We showcase the recent advances on four selected applications that explore 360°media. [ABSTRACT FROM AUTHOR]

Semantic segmentation is a difficult task in computer vision that have applications in many scenarios, often as a preprocessing step for a tool. Current solutions are based on Deep Neural Networks, which often require a large amount of data for learning a task. Aiming to alleviate the strenuous data-collecting/annotating labor, research fields have emerged in recent years. One of them is Meta-Learning, which tries to improve the generability of models to learn in a restricted amount of data. In this work, we extend a previous paper conducting a more extensive overview of the still under-explored problem of Few-Shot Weakly-supervised Semantic Segmentation. We refined the previous taxonomy and included the review of additional methods, including Few-Shot Segmentation methods that could be adapted to the weak supervision. The goal is to provide a simple organization of literature and highlight aspects observed in the current moment, and be a starting point to foment research on this problem with applications in areas like medical imaging, remote sensing, video segmentation, and others. [Display omitted] • A review of Non-Meta-Learning methods for weakly-supervised segmentation. • A review of Meta-Learning approaches for few-shot segmentation and classification. • A proposed taxonomy to Meta-Learning methods for few-shot semantic segmentation. • Limitations and future directions for few-shot weak-supervised segmentation. [ABSTRACT FROM AUTHOR]

Fractured object assembly is a fundamental problem across many disciplines such as digital archaeology, accident recovery, paleontology, cranio-facial and orthopaedic surgical repair, and many more. With advances in 3D acquisition devices and improved fidelity models of fractured pieces, a flurry of semi-automatic and automatic computational assembly solutions have been developed. These methods have the potential to reduce human labor in assembling fractured objects. The challenge is to balance the time efficiency with the fidelity of the reconstructed assembly. This paper introduces TAssembly , a data-driven approach to fractured object assembly. TAssembly leverages learned feature descriptors of the underlying objects. Its assembly pipeline seamlessly integrates multiple objective terms, including feature matching, rigid matching, and regularization for matching and stitching adjacent pieces. Many experimental results show that TAssembly significantly outperforms previous fractured object assembly methods. [Display omitted] • A data-driven method for 3D fracture bone assembly is proposed. • A latent shape descriptor space for matching features on fractured objects. • Using Stitching constraints to discourage self-penetration. • Feature extraction, feature matching, pairwise matching, consistent matching. [ABSTRACT FROM AUTHOR]

Sankey diagrams are widely used to visualize event sequence data. However, when the data volume is large, its readability is affected by dense edge crossings, excessive swing amplitude, and small crossover angles, while it is computationally intensive to obtain an optimal layout. In this paper, we propose NeatSankey, a balanced method that generates Sankey diagrams smoothly. It can be laid out quickly with good readability when Sankey diagrams are very complex. At the same time, to comprehensively evaluate the readability of Sankey diagrams, we use three evaluation metrics: crossing number, swing amplitude, and layout coverage. Firstly, we use a heuristic layout algorithm and a force-directed algorithm to adjust the node layout to minimize the edge crossings and swing amplitude with edge widths considered. Secondly, to better reduce the dense confusion caused by edge crossings, we introduce a edge bundling algorithm based on attribute similarity. We present three evaluations: a comprehensive comparison of our results with state-of-the-art techniques, user studies with thirty volunteers, and a case study of two datasets. Our evaluations demonstrate the effectiveness and practicability of the NeatSankey. [Display omitted] • We propose a novel two-step algorithm to enhance the readability of Sankey diagrams. • NeatSankey can generate Sankey diagrams smoothly within a limited time. • We can bundle edges with different widths and show the exact connections. • We design evaluations which can measure the readability of Sankey diagrams perfectly. • We conduct three studies to show the effectiveness and practicability of NeatSankey. [ABSTRACT FROM AUTHOR]

• An interactive design system to design sculpting styles and robotically fabricate on clay material. • A divide-and-conquer strategy for independent assigning of sculpting styles over complex geometries. • Multi-objective optimization for path planning of toolpath-driven robotic fabrication. • Sculpting style exploration and analysis inspired by painting brushstroke for robotic clay sculpting. [Display omitted] This paper presents an interactive design system that allows the user to create and fabricate stylized sculptures in water-based clay, using a standard 6-axis robot arm. This system facilitates the materialization of abstract design intentions into clay, through the algorithmic formulation of sculpting styles, the optimal path planning of the sculpting toolpaths, and a subtractive robotic fabrication process using customized tools. Unlike other precision-driven fabrication technologies, the authors embrace artistic uncertainty by conducting manual and robotic sculpting experiments and incorporating prominent parameters that affect the fabrication quality. The versatility of the described approach is demonstrated by designing a series of sculpting styles over a wide range of 3D models and robotically fabricating them in clay. Additionally, the paper explores various strategies for designing stylized robotic sculpting patterns by generating toolpaths informed by different techniques. [ABSTRACT FROM AUTHOR]

• The current 3D human body reconstruction technology requires a large number of 3D scanning equipment and a complex dynamic capture process, which significantly limits its rapid adoption in these application domains. The purpose of this article is to simplify the process of 3D reconstruction. • By combining deep learning methods, an end-to-end 3D human body reconstruction framework is proposed to achieve the goal of reconstructing a high-precision 3D human body model based on RGB images. [Display omitted] In recent years, with the improvement of artificial intelligence technology, it has become possible to reconstruct high-precision 3D human body models based on ordinary RGB images. The current 3D human body reconstruction technology requires complex external equipment to scan all angles of the human body, which is complicated to be implemented and cannot be popularized. In order to solve this problem, this paper applies deep learning models on reconstructing 3D human body based on monocular images. First of all, this paper uses Stacked Hourglass network to perform convolution operations on monocular images collected from different views. Then Multi-Layer Perceptrons (MLPs) are used to decode the encoded high-level images. The feature codes in the two views(main and side) are fused, and the interior and exterior points are classified by the fusion features, so as to obtain the corresponding 3D occupancy field. At last, the Marching Cube algorithm is used for 3D reconstruction with a specific threshold and then we use Laplace smoothing algorithm to remove artifacts. This paper proposes a dense sampling strategy based on the important joint points of the human body, which has a certain optimization effect on the realization of high-precision 3D reconstruction. The performance of the proposed scheme has been validated on the open source datasets, MGN dataset and the THuman dataset, provided by Tsinghua University. The proposed scheme can reconstruct features such as clothing folds, color textures, and facial details,and has great potential to be applied in different applications. [ABSTRACT FROM AUTHOR]

• The iterative adaptive depth estimation for learning-based MVS to avoid excessive computation on well-estimated regions during each refinement stage. • A location selection strategy based on geometric consistency to select locations where depth hypotheses are likely to be incorrect. • A depth candidate construction strategy for the selected location based on geometric information aggregation from multiple views. • The pixelwise depth estimation module which can estimate depth value for a single location independently and be utilized for sparse depth estimation. [Display omitted] In this paper, we propose a Multi-View Stereo (MVS) network which can perform high-quality depth estimation with low memory consumption. The proposed MVS network is constructed based on the pyramid architecture to gradually refine and upsample the depth map to the desired resolution. Instead of estimating depth hypotheses for all pixels in the depth map, our method only performs prediction at adaptively selected locations, alleviating excessive computation on well-estimated positions. To estimate depth hypotheses for sparse selected locations, we propose the lightweight pixelwise depth estimation module, which can estimate accurate depth value for each selected location independently. In this paper, we propose a Multi-View Stereo (MVS) network which can perform efficient high-resolution depth estimation with low memory consumption. Classical learning-based MVS approaches typically construct 3D cost volumes to regress depth information, making the output resolution rather limited as the memory consumption grows cubically with the input resolution. Although recent approaches have made significant progress in scalability by introducing the coarse-to-fine fashion or sequential cost map regularization, the memory consumption still grows quadratically with input resolution and is not friendly for commodity GPU. Observing that the surfaces of most objects in real world are locally smooth, we assume that most of the depth hypotheses upsampled from a well-estimated depth map are accurate. Based on the assumption, we propose a pyramid MVS network based on the adaptive depth estimation, which gradually refines and upsamples the depth map to the desired resolution. Instead of estimating depth hypotheses for all pixels in the depth map, our method only performs prediction at adaptively selected locations, alleviating excessive computation on well-estimated positions. To estimate depth hypotheses for sparse selected locations, we propose the lightweight pixelwise depth estimation network, which can estimate depth value for each selected location independently. Experiments demonstrate that our method can generate results comparable with the state-of-the-art learning-based methods while reconstructing more geometric details and consuming less GPU memory. [ABSTRACT FROM AUTHOR]

The B-spline technique is attractive to fit complex image outlines because of its freeform character. However, knot placement remains a difficult task. In this paper, the knots optimization problem is converted to an issue of feature points optimization, and a residuals-distribution-guided local optimization (RDG-LO) method is proposed to promote the quality of fitted curves. The RDG-LO algorithm states the moving direction and distance of candidate feature points according to the distribution of residuals. For a fixed number of feature points, the accuracy of the fitted curve is improved by adjusting the positions of feature points and the length of the fitted curve. Experimental results show that the fitted curve after optimization possesses advantages in both accuracy and simplicity. This algorithm is of great significance in the three-dimensional digital reconstruction of biomedical images. [Display omitted] • Using a B-spline curve to fit image outlines. • Optimizing the positions of feature points. • Improving the accuracy and simplicity of the fitted curve. [ABSTRACT FROM AUTHOR]

To meet the growing demand of rich 3D content, the Computer Graphics industry needs tools to automatically create procedural textures and materials from image exemplars. In this paper, we focus on the inverse procedural modeling of cellular stochastic structures , i.e., spatial distributions of repetitive, possibly blending similar shapes over a planar surface, resulting from stochastic processes. Such structures are very frequent in natural textures and materials, that exhibit random spatial variations. We represent cellular stochastic structures procedurally using thresholded Point Process Texture Basis Functions (PPTBFs). Previous approaches that learn PPTBF representations of structures maps solely rely on sampled data obtained by uniformly sampling the PPTBF parameter space, and parameter prediction based on these representations fails when the exemplars are visually too far from the training datasets. For the specific class of cellular stochastic structures, we propose to overcome this limitation by introducing a Cellular PPTBF, or C-PPTBF, defined with differentiable window and feature functions. Based on this procedural model, we present a fully differentiable pipeline and optimization procedure to automatically estimate the parameters of a C-PPTBF. We show that our method is efficient (between 2 and 5 min per exemplar) and yields more robust parameter prediction than the state-of-the-art DiffProxy approach. [Display omitted] • Inverse procedural modeling of cellular stochastic structures based on PPTBFs. • Differentiable window and feature functions that cover cellular stochastic structures. • Fully differentiable pipeline for parameters estimation from structure maps exemplars. • Training dataset size reduced by 51% in comparison to state-of-the-art DiffProxy approach. • Improved robustness w.r.t. state-of-the-art DiffProxy approach. [ABSTRACT FROM AUTHOR]

The existing 3D deep learning methods adopt either individual point-based features or local-neighboring voxel-based features, and demonstrate great potential for processing 3D data. However, the point-based models are inefficient due to the unordered nature of point clouds and the voxel-based models suffer from large information loss. Motivated by the success of recent point-voxel representation, such as PVCNN and DRINet, we propose a new convolutional neural network, called Multi Point-Voxel Convolution (MPVConv), for deep learning on point clouds. Integrating both the advantages of voxel and point-based methods, MPVConv can effectively increase the neighboring collection between point-based features and also promote independence among voxel-based features. Extensive experiments on benchmark datasets such as ShapeNet Part, S3DIS and KITTI for various tasks show that MPVConv improves the accuracy of the backbone (PointNet) by up to 36% , and achieves higher accuracy than the voxel-based model with up to 34 × speedups. In addition, MPVConv outperforms the state-of-the-art point-based models with up to 8 × speedups. Also, our MPVConv only needs 65% of the GPU memory required by the latest point-voxel-based model (DRINet). The source code of our method is attached in https://github.com/NWUzhouwei/MPVConv. [Display omitted] • In this paper, we propose Multi Point-Voxel Convolution (MPVConv) which takes the advantages of both the voxel- and point-based methods, and can work with different backbones for a wide range of 3D tasks. • Our MPVConv increases not only the neighboring collection for point-based features, but also the independence among voxel-based features. • 1×1×1 3D CNN is adopted in fusion module to strengthen the information association and nonlinearity between point-based and voxel-based features. • We replace the tri-linear interpolation with nearest-neighbor interpolation to reduce the computation cost of voxel feature interpolation. [ABSTRACT FROM AUTHOR]

Weathering phenomena are ubiquitous in urban environments, where it is easy to observe severely degraded old buildings as a result of water penetration. Despite being an important part of any realistic city, this kind of phenomenon has received little attention from the Computer Graphics community compared to stains resulting from biological or flow effects on the building exteriors. In this paper, we present physically-inspired deep weathering effects, where the penetration of humidity (i.e., water particles) and its interaction with a building's internal structural elements result in large, visible degradation effects. Our implementation is based on a particle-based propagation model for humidity propagation, coupled with a spring-based interaction simulation that allows chemical interactions, like the formation of rust, to deform and destroy a building's inner structure. To illustrate our methodology, we show a collection of deep degradation effects applied to urban models involving the creation of rust or of ice within walls. [Display omitted] • Physically-inspired simulation of the penetration of water particles in walls, dealing with multiple types of materials. • Approximate simulation of the interactions of water with the inner structures of the wall. • Expansion and deformation, such as water into ice, rusting, or plaster deformation. • Simulation of the formation of cracks inside the wall, leading to broken concrete, loose plaster, etc. • Cracks can accelerate previous steps if increased exposure (e.g., rain). [ABSTRACT FROM AUTHOR]

The use of social media has become so popular that people share photos every day on them. Automatic face recognition and tagging of people's photos have caused privacy preservation issues and some methods have been proposed for hiding the identity of presented people in these images. Blurring and blacking the face area, adding physical adversarial patches to the face, and adding adversarial masks are some proposed methods for this purpose. However, these methods particularly suffer from dissimilarity of the input and output images and inadequate performance in identity concealment from automatic face recognition (AFR) systems. In this paper, we propose the Generative Mask-guided Face Image Manipulation (GMFIM) model based on Generative Adversarial Networks (GANs) to apply imperceptible edits to the input face image to preserve the identity of the person in the image. Our model consists of a face mask module, a GAN-based optimization module, and a merge module. Different criteria are considered in the objective function of the optimization step to produce high-quality images that are as similar as possible to the input image while they cannot be recognized by AFR systems. The results of the experiments on different datasets show that our model provides promising results in terms of the quality of the generated images and the identity concealment performance. [Display omitted] • A mask-guided GAN-based model (GMFIM) is proposed for deidentification of face images. • An optimization module based on two loss functions is proposed for hiding the identity of images. • A mask-guided model is adopted for separating the face area from the background. • A Fourier transform based model is used for merging the modified face with the background. • Qualitative and quantitative experiments are performed to evaluate and analyze the proposed model. [ABSTRACT FROM AUTHOR]

Integer Grid Maps (IGMs) are a class of mappings characterized by integer isolines that align up to unit translations and rotations of multiples of 90 degrees. They are widely used in the context of remeshing, to lay a quadrilateral grid onto the mapped surface. The presence of both discrete and continuous degrees of freedom makes the computation of IGMs extremely challenging. In particular, solving for all degrees of freedom altogether leads to a mixed-integer problem that is known to be NP-Hard. Such a problem can only be solved heuristically, occasionally failing to produce a valid quadrilateral mesh. In this paper we propose a simple topological construction that allows to reduce the problem of computing a valid IGM to the one of mapping a topological disk to a convex domain. This is a much easier problem to deal with, because it completely removes the integer constraints, permitting to obtain a provably injective parameterization that is guaranteed to incorporate all the correct integer transitions with a simple linear solve. Not only the proposed algorithm is easy to implement, but it is also independent from costly numerical solvers that are unavoidable in existing quadmeshing pipelines, preventing their exploitation in open source or low-budget projects. Despite provably correct, the so generated maps contain a considerable amount of geometric distortion and a poor quad connectivity, making this technique more suitable for a robust initialization rather than for the computation of an application-ready IGM. In the article we present the details of our construction, also analyzing its geometric and topological properties. [Display omitted] • A method to compute a provably injective integer grid map is presented. • A method to enforce the normal form in a canonical polygonal schema is presented. • Future perspectives towards the generalization to higher dimensional integer grid maps are discussed. [ABSTRACT FROM AUTHOR]

For over two decades, the OpenGL API provided users with the means for implementing versatile, feature-rich, and portable real-time graphics applications. Consequently, it has been widely adopted by practitioners and educators alike and is deeply ingrained in many curricula that teach real-time graphics for higher education. Over the years, the architecture of graphics processing units (GPUs) incrementally diverged from OpenGL's conceptual design. The more recently introduced Vulkan API provides a more modern, fine-grained approach for interfacing with the GPU, which allows a high level of controllability and, thereby, deep insights into the inner workings of modern GPUs. This property makes the Vulkan API especially well suitable for teaching graphics programming in university education, where fundamental knowledge shall be conveyed. Hence, it stands to reason that educators who have their students' best interests at heart should provide them with corresponding lecture material. However, Vulkan is notoriously verbose and rather challenging for first-time users, thus transitioning to this new API bears a considerable risk of failing to achieve expected teaching goals. In this paper, we document our experiences after teaching Vulkan in both introductory and advanced graphics courses side-by-side with conventional OpenGL. A collection of surveys enables us to draw conclusions about perceived workload, difficulty, and students' acceptance of either approach. In doing so, we identify suitable conditions and recommendations for teaching Vulkan to both undergraduate and graduate students. [Display omitted] • How to transition introductory graphics courses from OpenGL to Vulkan. • Programming framework suitable for introductory graphics courses. • Student questionnaire evaluations about the transition to Vulkan. • Vulkan usage in advances graphics courses. • Programming framework suitable for teaching in advanced graphics courses and research. [ABSTRACT FROM AUTHOR]

Electrotactile feedback is a convenient tactile rendering method thanks to reduced form factor and power consumption. Yet, its usage in immersive virtual reality has been scarcely addressed. This paper explores how electrotactile feedback could be used to enhance contact information for mid-air interactions in virtual reality. We propose an electrotactile rendering method which modulates the perceived intensity of the electrotactile stimuli according to the interpenetration distance between the user's finger and the virtual surface. In a first user study (N=21), we assessed the performance of our method against visual interpenetration feedback and no-feedback. Contact precision and accuracy were significantly improved when using interpenetration feedback. The results also showed that the calibration of the electrotactile stimuli was key, which motivated a second study exploring how the calibration procedure could be improved. In a second study (N=16), we compared two calibration techniques: a non-VR keyboard and a VR direct interaction method. While the two methods provided comparable usability and calibration accuracy, the VR method was significantly faster. These results pave the way for the usage of electrotactile feedback as an efficient alternative to visual feedback for enhancing contact information in virtual reality. [Display omitted] • Electrotactile improves contact's accuracy and precision when touching virtual objects • Combination of visual and electrotactile feedback improves contact rendering even more • Calibration of electrotactile can be done faster using a virtual reality interaction. [ABSTRACT FROM AUTHOR]

For engineering applications, an innovative design can be developed by reusing and modifying existing models with similar features and manufacturing properties, and accurately searching CAD models has become a valuable knowledge acquisition technique in product design. Although there are many retrieval technologies, most methods focus on the global shape of models. In this paper, a CAD model retrieval framework based on local feature segmentation is proposed, offering users a free-form way to choose any subpart as a query from a given model. First, a novel model segmentation method based on the vertex-neighbor extension is proposed to divide a target CAD model into overlapping local features. The chosen subpart with arbitrary boundaries is also successfully described for retrieval by combining local features. Then, a composite descriptor considering both shapes and attribute characteristics is established, which is proved effective for distinguishing local features. Finally, the problem of matching free-form queries to CAD models in the dataset is transformed into descriptor set measurement and is implemented using the bag-of-word algorithm. The proposed model retrieval approach outperforms many existing approaches in matching performance and provides a user-friendly query mode. [Display omitted] • Solve the partial CAD models retrieval (B-rep) to support engineering reuse. • Introduce a innovative model segmentation method to generate local features. • Apply the bag-of-word algorithm to measure the similarity. • Allow arbitrary queries acquiring from free-form interaction with the CAD system. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel approach for sampling indirect illumination based on bidirectional path tracing and world space resampling reservoir. We cache light subpaths with high contribution in a world space grid, allowing for sampling at arbitrary vertices along the eye path. In order to find optimal samples, we continuously update the reservoirs stored in each cell according to the resampling principles. Experiments have shown that our algorithm achieves noticeable noise reduction in scenes lit by indirect light compared with path tracing, at the cost of additional time and memory consumption, while still maintaining interactive frame rates. In some special cases, our method even shows better results than a plain bidirectional path tracer, in terms of both image quality and performance. [Display omitted] • Path tracing (PT) has a poor performance at scenes lit mainly by indirect light. • Bidirectional path tracing (BDPT) is more efficient than PT but too resource-intensive. • World space resampling reservoir makes the sampling of light subpaths more efficient. • Our method achieves remarkable noise reduction compared with PT and BDPT. • Our method runs at an interactive frame rate. [ABSTRACT FROM AUTHOR]

Person image generation is a challenging task aimed to transfer the person of the source image from a source pose to a target pose while preserving its style. In this paper, we proposed a Generative Adversarial Network based on Decoupled Semantic Attention Transfer (DSAT-GAN), focusing on that local semantic representations of different image styles and contents cannot be accurately decoupled and transferred. This architecture employs a novel Multi-scale Semantic Mapping Generation Network (Ms-SMGN), driven by two network modules with different semantic attention mechanism, aiming to accurately align and transfer the representations of local semantics at different spatial scales. Then, a channel-separated convolution is applied in the encoding networks instead of the traditional channel fully-connected operation, which reduces computational complexity while realizing channel semantic decoupling. Moreover, a Gram matrix-based global style loss is introduced to further enhance the consistency of high-level semantic between generated and target images. Experiments on Market-1501 and DeepFashion datasets show that DSAT-GAN has superior performance compared with other recent baselines. Additionally, this architecture can be extended to the data enhancement scenes to significantly improve the accuracy of person Re-identification. [Display omitted] • Our model alleviates the loss of semantic features of complex alignment regions. • Encoders learn feature vectors with channel-separated convolution and attention. • Decoders can fuse and decode feature vectors and generate integrated person details. • Global style loss constrain the high-level semantic of generated person images. • The experimental results on the datasets demonstrate the superiority of our model. [ABSTRACT FROM AUTHOR]

• Our sketching tasks is unique and has not been used in previous research into immersive 3D sketching despite calls for more innovative methods. • Our observations show that immersive 3D sketching provides unique visualization opportunities while also supporting behaviors that are common in paper-based sketching as well as those found in gesturing, physical modeling and 3D CAD. • This unique combination enables users to engage in visual thinking and visual communication activities in ways that cannot presently be achieved with any other single representation technique. • It is widely accepted that stroke-based immersive 3D sketching is insufficient for the representation of 3D shapes, based on the assumption that precision is more important than flexibility. We show this assumption does not hold for visual thinking and visual communication. • Our findings suggest that the value of stroke-based VR sketching is in its capacity to challenge participants to think more carefully about the representation and visual communication of 3D shapes. Research Highlights Immersive 3D sketching engages users in visual thinking and communication in ways not possible with any other single methodImmersive 3D sketching offers unique visualization opportunities, combining those of traditional physical and digital mediaImmersive 3D sketching challenges users to think carefully about the representation and communication of shapes and spacesImmersive 3D sketching is valued for challenging assumptions that accuracy is more important than flexibility for these tools Paper-based (PB) sketching involves the challenge of representing three-dimensional (3D) shapes on two-dimensional (2D) surfaces. The recent generation of virtual reality (VR) sketching tools offer a way to overcome this challenge. These immersive 3D sketching environments permit the rapid construction of freehand stroke-based 3D models in 3D space while replicating the immediate experience of PB sketching. To explore the potential advantages of VR sketching in visual thinking and visual communication, we conducted investigations with sixteen architectural students engaged in PB and VR sketching tasks. We observed their visualization behavior during VR sketching and their behavior in transitioning between PB and VR sketching. The participants experiences of the two media were also recorded in semi-structured interviews and questionnaires. Our observations show that immersive 3D sketching is a unique form of visual representation that facilitates the rapid and flexible creation of large and detailed (but inaccurate) 3D computer models. It is a multimodal medium that supports visual thinking and communication behaviors associated with PB sketching, CAD modeling, physical model-making and gesturing, all within the same space. This unique combination enables users to engage in visual thinking and visual communication activities in ways that cannot presently be achieved with any other single representation technique. [ABSTRACT FROM AUTHOR]

• Framework for analysis of colon data. • Automatic extraction of morphological data. • Web-based tool for visual exploration of 2D and 3D data. • Evaluation. Our paper presents a complete system for the data extraction and visual exploration of colon contents and colonmorphological data. The system uses previous research in colon segmentation of MRI images and contributes with anew module for the extraction of morphological data (MGeom), and two modules for the visual exploratory analysis ofthe extracted data (MVisPlot3D and MVisPlot2D).The modules in orange are the ones presented in this paper. In order to extract the colon data (contents andmorphology) we have two different MRI-based sources: T2 and T1-FS images. The two leftmost modules are used tosegment the T2 and T1-FS images. The second column of modules shows the pieces of our framework intended toextract morphological (top) and colon contents (bottom) data. The rightmost modules are used for the visualinspection of the extracted data, as well as other data that we can have as input (e.g. motility data). Computerized Tomography (CT) and, more recently, Magnetic Resonance Imaging (MRI) have become the state-of-the art techniques for morpho-volumetric analysis of abdominal cavities. Due to its constant motility, the colon is an organ difficult to analyze. Unfortunately, CT's radiative nature makes it only indicated for patients with important disorders. Lately, acquisition techniques that rely on the use of MRI have matured enough to enable the analysis of colon data. This allows gathering data of patients without preparation (i.e. administration of drugs or contrast agents), and incorporating data of patients with non life-threatening diseases and healthy subjects to databases. In this paper we present an end-to-end framework that comprises all the steps to extract colon content and morphology data coupled with a web-based visualization tool that facilitates the visual exploration of such data. We also introduce the set of tools for the extraction of morphological data, and a detailed description of a specifically-designed interactive tool that facilitates a visual comparison of numerical variables within a set of patients, as well as a detailed inspection of an individual. Our prototype was evaluated by domain experts, which showed that our visual approach may reduce the costly process of colon data analysis. As a result, physicians have been able to get new insights on the effects of diets, and also to obtain a better understanding on the motility of the colon. [ABSTRACT FROM AUTHOR]

Skeleton-based action sequences are widely used for human behaviour understanding due to their compact characteristics. Most existing work designed Graph Convolutional Networks and integrated multiple input channels rather than the original motion sequence to improve the final performance. However, few of them have been reported on the detailed effects of such multiple input channels. In contrast to them, we systemically study the impact of different input channels and construct a more efficient GCN framework. We have identified the complementary effect between the local frame channel and global sequence channel, which is essential to improve the action recognition accuracy. By coupling local frame and global sequence information with a classical spatial–temporal graph neural network, e.g. MS-G3D, it achieves competitive performance compared with SOTA methods on challenging benchmark datasets. Related code would be available at https://github.com/movearbitrarily/multi-stream. • In this paper, we propose an effective multi-channel mechanism that exploiting global and local complementary information, which could significantly improve GCN models' performance on skeleton based action recognition. • We conduct comprehensive evaluations on two large scale benchmark datasets and achieved a competitive result compared with SOTA approaches. • We show that the performance of GCN models could be improved in a simple way. • Notably, the proposed multi-channel mechanism is generic and robust to seamlessly work with any existing GCN framework, which is essential that the reasonable performance could be achieved in a simple way. [ABSTRACT FROM AUTHOR]

This paper proposes a real-time fluid rendering method based on the screen space rendering scheme for particle-based fluid simulation. Our method applies anisotropic transformations to the point sprites to stretch the point sprites along appropriate axes, obtaining smooth fluid surfaces based on the weighted principal components analysis of the particle distribution. Then we combine the processed anisotropic point sprite information with popular screen space filters like curvature flow and narrow-range filters to process the depth information. Experiments show that the proposed method can efficiently resolve the issues of jagged edges and unevenness on the surface that existed in previous methods while preserving sharp high-frequency details. • Improving quality for highly efficient rendering of fluid simulations. • Resolving the issues of jagged edges and unevenness on the fluid surface. • Efficiently preserving sharp high-frequency details. • Easy to be implemented with the state-of-the-art smoothing filters. [ABSTRACT FROM AUTHOR]

In this paper, we present FaceTuneGAN, a new 3D face model representation decomposing and encoding separately facial identity and facial expression. We propose a first adaptation of image-to-image translation networks, that have successfully been used in the 2D domain, to 3D face geometry. Leveraging recently released large face scan databases (FaceScape and CoMA), a neural network has been trained to decouple factors of variations with a better knowledge of the face, enabling facial expressions transfer and neutralization of expressive faces. Specifically, we design an adversarial architecture adapting the base architecture of FUNIT and using SpiralNet++ for our convolutional and sampling operations. Applied on these two datasets, FaceTuneGAN has a better identity decomposition and face neutralization than state-of-the-art techniques. It also outperforms classical deformation transfer approach by predicting blendshapes closer to ground-truth data and with less of undesired artifacts due to too different facial morphologies between source and target. [Display omitted] • 3D face model representation. • Morphology aware expression transfer. • Style based 3D facial generative model. • Semi-supervised learning of 3D identity and expression separation. [ABSTRACT FROM AUTHOR]

Robotic-Assisted Surgery (RAS) is beginning to unlock its potential. However, despite the latest advances in RAS, the steep learning curve of RAS devices remains a problem. A common teaching resource in surgery is the use of videos of previous procedures, which in RAS are almost always stereoscopic. It is important to be able to add virtual annotations onto these videos so that certain elements of the surgical process are tracked and highlighted during the teaching session. Including virtual annotations in stereoscopic videos turns them into Mixed Reality (MR) experiences, in which tissues, tools and procedures are better observed. However, an MR-based annotation of objects requires tracking and some kind of depth estimation. For this reason, this paper proposes a real-time hybrid tracking–matching method for performing virtual annotations on RAS videos. The proposed method is hybrid because it combines tracking and stereo matching, avoiding the need to calculate the real depth of the pixels. The method was tested with six different state-of-the-art trackers and assessed with videos of a sigmoidectomy of a sigma neoplasia, performed with a Da Vinci® X surgical system. Objective assessment metrics are proposed, presented and calculated for the different solutions. The results show that the method can successfully annotate RAS videos in real-time. Of all the trackers tested for the presented method, the CSRT (Channel and Spatial Reliability Tracking) tracker seems to be the most reliable and robust in terms of tracking capabilities. In addition, in the absence of an absolute ground truth, an assessment with a domain expert using a novel continuous-rating method with an Oculus Quest 2 Virtual Reality device was performed, showing that the depth perception of the virtual annotations is good, despite the fact that no absolute depth values are calculated. [Display omitted] • Enhancing videos of robotic surgery with virtual annotations is useful for training. • Tracking and depth estimation often required for annotation of stereoscopic videos. • A real-time tracking–matching framework is proposed; true pixel depth not necessary. • Validation with objective metrics and a domain expert using a continuous rating. • CSRT-based method provides suitable tracking and perception of virtual annotations. [ABSTRACT FROM AUTHOR]

The eXtended Position Based Dynamics algorithm (XPBD) enables unified simulation of various materials from fluids to both elastic solids and stiff solids. In particular, finite element based neo-Hookean models can simulate near incompressible materials by means of a decoupled compliant constraint formulation. Due to XPBD's reliance on local constraint projections in the solver loop, its computational nature lends itself to parallelization by means of graph coloring algorithms used to determine partitions of independent constraints which can be solved simultaneously. However, minimal graph coloring is bounded from below by the maximum valence of the finite element mesh, thus hindering parallelization opportunities. In this paper, we propose a novel graph clustering approach on the constraint graph which groups highly dependent constraints into supernodes. By applying graph coloring on the supernodal constraint graph, we are able to significantly reduce the number of partitions, thus enhancing parallelization of the solver. Furthermore, we accelerate convergence of the neo-Hookean XPBD solver by a coupled constraint formulation, resulting in enhanced stability and efficiency compared to previous approaches. • Constraint graph clustering improves partitioning based parallelism for soft body solvers. • Blocked constraint projection accelerates convergence. • Order of magnitude speedup compared to baseline simulation techniques. [ABSTRACT FROM AUTHOR]

Despite the wide availability of large motion databases, and recent advances in exploiting them to create tailored contents, artists still spend tremendous efforts in editing and adapting existing character animations to new contexts. This motion editing problem is a well-addressed research topic. Yet only a few approaches have considered the influence of the camera angle, and the resulting visual features it yields, as a mean to control and warp a character animation. This paper proposes the design of viewpoint-dependent motion warping units that perform subtle updates on animations through the specification of visual motion features such as visibility, or spatial extent. We express the problem as a specific case of visual servoing, where the warping of a given character motion is regulated by a number of visual features to enforce. Results demonstrate the relevance of the approach for different motion editing tasks and its potential to empower virtual characters with attention-aware communication abilities. [Display omitted] • A procedural animation technique to adapt the motion of virtual characters. • Performing the analysis of low-level perception features from an observer point of view. • Increasing the realism of interaction and the expressive capability of digital avatars. • Increasing social presence in virtual reality applications. • Inspired by robotics' visual servoing, and applied to virtual characters animation. [ABSTRACT FROM AUTHOR]

In this paper, we present a novel real-time physical interaction system that allows users to throw a virtual ball without using an intermediary device such as a controller. A dataset of throwing motions was captured from six actors, from which ground truth Point of Release (PoR) frames were calculated. We trained a PoR prediction model using motion features extracted from arm joints, and developed a detection algorithm to predict the PoR of a throwing motion in real-time. Evaluation of the system with pre-recorded throwing motion data results in detection errors of less than 50 ms. Qualitative results from six users of the real-time system in VR indicate that the task of throwing without a controller was very natural, although the system performed better for underarm than overarm throws. Finally, we report the relative importance of different joints and motion features for the PoR prediction task. [Display omitted] • A corpus of 1679 throwing/catching motions annotated with Point of Release (PoR). • A PoR prediction model using motion features extracted from arm joints in real-time. • A real-time VR system for throwing interactions using only the user's arm motion. • A ranking of 15 joint and motion feature combinations for PoR prediction. [ABSTRACT FROM AUTHOR]

Developable surfaces are widely used in Computer-Aided Design (CAD), computer graphics, architecture geometry and manufacturing. In this paper, we present an approach to segmentation and approximation of an input triangular mesh with developable patches. First, exact developable regions are detected and extracted from the input mesh. Then the non-developable region is further segmented and approximated also by developable patches. The parametric equations for all obtained developable patches are simultaneously computed, whose iso-parameter curves can serve directly as the milling paths in CNC machining. Examples and comparisons with existing works are provided. [Display omitted] • A novel method for segmenting and approximating an input mesh with developable patches is proposed. • The exact parametric equations for all segment patches are computed. • A new linearity and smoothness measures to detect curve-like features on Gauss Images is provided. [ABSTRACT FROM AUTHOR]

The last two decades have exhibited a profound transformation of traditional urban mobility patterns partly due to the exponential growth in both number and popularity of public bicycle sharing systems (BSS). Analysis and visualization of the data generated by BSSs have become of special interest to municipalities to evaluate the effect of their mobility programs and offer integrated urban mobility solutions. In this paper, we present a visualization system that aims to assist city officials from small and medium cities in their decision-making process with an intuitive representation of BSS' data. It has been developed, tested, and evaluated together with officials and domain experts from the city of Logroño (Spain). Our tool presents usage information with different time granularities (yearly, monthly, weekly, or seasonally), shows traffic flows between stations, and provides an in-depth breakdown of users' data such as their registered address, traveled distance, or gender-based patterns. [Display omitted] • Web-based exploratory tool to analyze usage trends of bike sharing systems. • Analysis tool to support mobility policies planning. • A report-mode that enables generation of data-based detailed reports on demand. • Scalability to other bike sharing systems with similar needs. [ABSTRACT FROM AUTHOR]

In this paper, we present a structure simplification framework for planar all-quad meshes with open boundaries. Our simplification framework can handle quad meshes with complex structures (e.g., quad meshes obtained via Catmull–Clark subdivision of the triangle meshes) to produce simpler meshes while preserving the boundary features. To achieve that, we introduce a set of separatrix-based semi-global operations and combine them with existing local operations to develop a new simplification framework. Additionally, we organize and order the individual simplification operations into groups and employ ranking strategies for each group to sort these operations to produce quad meshes with better quality and simpler structure. We provide a comprehensive evaluation of our framework using different input parameters on a number of representative planar quad meshes with various boundary configurations. To demonstrate the advantages of our method, we compare it with a few existing frameworks. Our comparison shows that our simplification framework usually produces simpler structure with faster computation than the state-of-the-art methods. [Display omitted] • Structure simplification used to optimize structural complexity of planar quad meshes. • Existing local and global simplification may produce sub-optimal results. • Semi-global simplification utilizes separatrices to perform simplification. • A new framework combines, organizes, and ranks semi-global and local operations. • Results show that the new framework outperforms existing simplification methods. [ABSTRACT FROM AUTHOR]

• This work proposes a novel computational framework for design of planar gears. • This is achieved by iterative conjugation which is shown to converge within two steps. • A robust computational framework for planar sweeps is proposed for conjugation. Gears play a pivotal role in machine design. This paper proposes an algorithm to simplify the shapes of planar gears. This is achieved via iterative conjugation, using precise algebraic sweeps. The notion of shape simplification is introduced in a mathematically rigorous manner and it is shown that the conjugation process converges, yielding a pair of meshing gears that follow the desired motion. Simplified gear shapes may lead to improved mechanical characteristics and reduction in manufacturing costs. The generality of algebraic sweeps allows precise design of gears with freeform shapes and non-uniform motion transmission. Moreover, the computational framework proposed in this paper is versatile, with applications beyond gear design. A variety of examples from an implementation of our algorithm, that offers topological guarantees, are presented, which demonstrate the robustness and efficacy of our approach. [ABSTRACT FROM AUTHOR]

• In this paper, we present a deep learning approach for cross-domain retrieving 3D shape and 2D sketch image. • We propose a new Cycle CNNs, which can construct the cross-domain mapping between the feature space of 2D sketches and the one of 3D shapes. • The core of our method is that the mapping relationship can be directly generated through the proposed Cycle CNNs, without explicit common feature space construction. • Our proposed Cycle CNNs works well on both sketch-based shape retrieval and shape-based sketch retrieval. In this paper, we present a deep learning approach for cross-domain retrieval of 3D shape and 2D sketch image. Cross-domain retrieval has received significant attention to flexibly find information across different modalities of data. Effective measuring the similarity between different modalities of data is the key of cross-domain retrieval. Different modalities such as shape and sketch have imbalanced and complementary relationships, which contain unequal amount of information when describing the same semantics. Existing methods based on deep learning networks mostly construct one common space for different modalities, and these nets usually loss exclusive modality-specific characteristics. To address this problem, we propose a novel Cycle CNNs to estimate the cross-domain mapping between the space of 3D shape descriptors and the one of 2D sketch features. First, we employ the existing networks to construct independent feature spaces for each modality. For each feature space, modality-specific properties within one modality are fully exploited. Next, we use the designed Cycle CNNs to learn the mapping function between different feature spaces. This network can capture the mapping relationship between 3D shape feature space and 2D sketch feature domain. Finally, we use the explored mapping between the feature spaces of different modalities to perform cross-domain retrieval. We demonstrate a variety of promising results, where our method achieves better retrieval accuracy than existing state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

• An integrated visualization of scholarly data can support the academic reviewer search process. • The visualization of scholarly data helps to avoid conflicts of interest and to build a fairly distributed pool of reviewers. • A well-combined visualization of citations and co-authorship relations only, can reduce the need for complicated content analysis techniques. • The platform evaluation with members from the Computer Graphics community demonstrated the improvement on the traditional process for searching reviewers. • The evaluation confirmed that users were able to get acquainted with the system with a very limited training. We propose ReviewerNet, an online, interactive visualization system aimed to improve the reviewer selection process in the academic domain. Given a paper submitted for publication, we assume that good candidate reviewers can be chosen among the authors of a small set of pertinent papers; ReviewerNet supports the construction of such set of papers, by visualizing and exploring a literature citation network. The system helps journal editors and Program Committee members to select reviewers that do not have any conflict-of-interest and are representative of different research groups, by visualising the careers and co-authorship relations of candidate reviewers. The system is publicly available, and is demonstrated in the field of Computer Graphics. [ABSTRACT FROM AUTHOR]

• Our survey investigated real locomotion for HMD-based Virtual Reality. • Our focus is on the interaction techniques - the combination of devices, user's actions and system's feedback. • We categorize the resulting real locomotion interaction techniques as Unmediated, Warnings, reorientation/resetting, Scaling, Redirection, and Dynamic VE. • We describe each category with different parameters that distinguish individual interaction techniques. Locomotion is a fundamental activity in Virtual Reality (VR) and has been the focus of a large body of research since the implementation of the first systems. In real locomotion techniques, users physically move in the real world to affect movement in the Virtual Environment (VE). Real locomotion has been found to perform better than other forms of locomotion for many tasks. To overcome the challenges imposed by restricted physical space, researchers have devised ingenious interaction techniques for real locomotion. In this paper, we present a survey of interaction techniques for real locomotion in VR. Our analysis is focused on the interaction techniques – the combination of devices, user's actions, and system's responses. We categorize interaction techniques for real locomotion in VR into: unmediated, warnings, reorientation/resetting, scaling, redirection, and dynamic VE. These categories represent fundamentally different approaches to real locomotion and user action feedback. We further characterize techniques in each of these categories according to category-specific parameters. Finally, it is important to state that this paper was developed with the aim of helping newcomers to the field to understand and implement the techniques here presented. [ABSTRACT FROM AUTHOR]

• We present a novel particle resampling method for surface reconstruction of a liquid. • The first adaptive sampling tailored to level-sets defined by the boundary particles. • Our adaptive particle resampling preserves small and thin features of a liquid. • A new quality metric to evaluate the effectiveness of the resampling methods. In this paper, we present a novel adaptive particle resampling method tailored for surface reconstruction of level-sets defined by the boundary particles from a particle-based liquid simulation. The proposed approach is simple and easy to implement, and only requires the positions of the particles to identify and refine regions with small and thin fluid features accurately. The method comprises four main stages: boundary detection, feature classification, particle refinement, and surface reconstruction. For each simulation frame, firstly the free-surface particles are captured through a boundary detection method. Then, the boundary particles are classified and labeled according to the deformation and the stretching of the free-surface computed from the Principal Component Analysis (PCA) of the particle positions. The particles placed at feature regions are then refined according to their feature classification. Finally, we extract the free-surface of the zero level-set defined by the resampled boundary particles and its normals. In order to render the free-surface, we demonstrate how the traditional methods of surface fitting in Computer Graphics and Computational Physics literature can benefit from the proposed resampling method. Furthermore, the results shown in the paper attest the effectiveness and robustness of our method when compared to state-of-the-art adaptive particle resampling techniques. [ABSTRACT FROM AUTHOR]